The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run() in scripts/check_env.py and tests/test_scripts.py. However, these are legitimate uses for repository inspection (e.g., git rev-parse) and unit testing. The run_cmd function in check_env.py is explicitly designed for read-only inventory and uses check=False with timeout protection.
[DYNAMIC_CONTEXT_INJECTION]: The skill documents and manages Claude Code SessionStart hooks in scripts/init_session_start_hook.py. The implementation is defensive: it uses atomic writes, validates guide paths against shell-sensitive characters, and restricts injection to a simple echo command for context nudging.
[DATA_EXPOSURE]: scripts/check_env.py and scripts/sanitize_history.sh search for the presence of credentials (like .env or API key patterns) but explicitly avoid reading or printing secret values, adhering to security best practices.
[INDIRECT_PROMPT_INJECTION]: The skill processes repository content (README, manifests, etc.) to infer setup steps. While this is an ingestion surface, the instructions emphasize evidence-driven diagnostics and require explicit user approval for destructive actions, mitigating the risk of following malicious instructions embedded in a repository.

auto-repo-setup